Encoding-pattern similarity did not predict recall transitions beyond serial position in 44 SEEG participants

We tested whether word-locked SEEG band-power similarity at encoding predicts the next recalled word beyond serial-position structure in 54 participants from OpenNeuro ds004789 (FR1, session 0); 44 participants met the planned threshold of at least 15 transitions. The preregistered primary statistic was the difference (D) between the percentile rank of the true next word’s similarity and a same-position, other-list swap null. For all bipolar pairs and five bands, P1 was negative (D = -0.033, 95% CI [-0.049, -0.016], Holm p = 0.003), opposite to the predicted direction, so the hypothesis was not supported. The later-recalled-candidates test (S1), which avoids a recall-status mismatch, was near zero (D = -0.003, 95% CI [-0.029, 0.025], p = 0.83). A post hoc analysis without recalled-flag regression changed P1 to +0.015 (95% CI [-0.001, 0.031], p = 0.09); the negative primary result therefore depends on that residualization and was interpreted as an artifact, while the alternative estimate may include subsequent-memory similarity and is not a clean context estimate. The analysis used 0–1.6 s log band power, all bipolar pairs, and session 0 only; it did not control semantic similarity or select regions. Source hashes, URLs, methods, code, and results are preserved; the exact release of the live-mirror event files is uncertain.

SEEG · free recall · memory · temporal context · band power · serial position · recall transitions · pattern similarity